A Neuromorphic Model to Match the Spiking Activity of Merkel Mechanoreceptors With Biomimetic Tactile Sensors for Bioengineering Applications

A major challenge in upper limb neuroprostheses is to reproduce the tactile feedback as the one provided by the spiking activity of human mechanoreceptors. In this paper, we aim at emulating the firing behavior of Merkel mechanoreceptors, innervated by slowly adapting type I (SA1) receptors, in silico by means of a custom implementation of the Izhikevich spiking neuron model with a porting function of the sensors output. We matched the neuron model output with the sustained firing observed in neurophysiological experiments in response to constant stimuli to the skin. We compared different input transformation functions to find the proper trade-off between agreement with biological spiking activity and model leanness. We identified a porting function that converts the output of physical sensors into the input to artificial spiking models of mechanoreceptors. The porting function was then validated by comparing the firing rate-indentation curve of real mechanoreceptors with the one obtained from an MEMS-based biomimetic tactile sensor platform. From the analysis of the adjusted-residual variance, the two curves result coherent. Therefore, having applied the calibration inverse function to a sensor output, the proposed porting function allows obtaining the proper input to an artificial neuron model, enabling the generation of neuromorphic signals.

I. Introduction

The objective of reproducing the dynamics of the human somatosensory system is typically pursued for two main bioengineering applications: neuroprosthetics and neurorobotics. Regarding the neuroprosthetics field, users of current artificial hand prostheses report the complete absence of sensory feedback as a relevant drawback of those systems [1], [2]. Tactile feedback restoration would then lead to an improved capability of manipulation, and possibly to a better embodiment of the prosthesis; such restoration could be possible thanks to a signal closely mimicking the physiological way in which the tactile peripheral system encodes information. In the neurorobotics field, the interest in mimicking tactile systems lies on the idea of building models which reproduce the advantageous features of natural sensory systems, as the reduction of the computational cost [3]. The developed artefacts, in turn, can be integrated for use in application domains such as surgical, industrial or field robotics, as well as used to test scientific hypotheses [4], [5].